benchmark mode specifies nb of threads with -v

[Cyan4973]

Now that the benchmark mode uses the same default nb of threads as normal CLI compression, it's no longer obvious how many threads are being used in benchmark. This can lead to incorrect conclusions, when comparing the benchmark result with earlier versions, which always use 1 thread by default.

With this PR, this information is displayed if the benchmark mode -b is combined with verbose output -v.

Example :

zstd -bv
Benchmarking at level 3 using 2 threads
 3#Lorem ipsum       :  10000000 ->   2983345 (x3.352),  205.2 MB/s,  866.3 MB/s

./zstd -b1e3v -S dirCalgary/b*    
Benchmarking 3 files from level 1 to 3 using 2 threads
 1#bib               :    111261 ->     40381 (x2.755),  266.2 MB/s, 1032.0 MB/s
 2#bib               :    111261 ->     39081 (x2.847),  243.3 MB/s   861.6 MB/s
 3#bib               :    111261 ->     37034 (x3.004),  196.4 MB/s   929.7 MB/s
 1#book1             :    768771 ->    349758 (x2.198),  161.5 MB/s, 1067.2 MB/s
 2#book1             :    768771 ->    317887 (x2.418),  139.5 MB/s,  824.2 MB/s
 3#book1             :    768771 ->    305288 (x2.518),  116.3 MB/s,  745.0 MB/s
 1#book2             :    610856 ->    228144 (x2.678),  238.9 MB/s, 1019.8 MB/s
 2#book2             :    610856 ->    215190 (x2.839),  182.4 MB/s,  860.3 MB/s
 3#book2             :    610856 ->    203784 (x2.998),  150.9 MB/s,  849.8 MB/s

compared to dev :

zstd -bv                                   
 3#Lorem ipsum       :  10000000 ->   2983345 (x3.352),  201.4 MB/s,  866.1 MB/s

zstd -b1e3v -S dirCalgary/b* 
Benchmarking levels from 1 to 3
Benchmarking /home/cyan/dev/bench/dirCalgary/bib 
 1#bib               :    111261 ->     40381 (x2.755),  267.1 MB/s, 1029.1 MB/s
 2#bib               :    111261 ->     39081 (x2.847),  245.2 MB/s   863.5 MB/s
 3#bib               :    111261 ->     37034 (x3.004),  194.7 MB/s   925.5 MB/s
Benchmarking /home/cyan/dev/bench/dirCalgary/book1 
 1#book1             :    768771 ->    349758 (x2.198),  204.3 MB/s, 1057.3 MB/s
 2#book1             :    768771 ->    317887 (x2.418),  133.5 MB/s,  822.4 MB/s
 3#book1             :    768771 ->    305288 (x2.518),  113.9 MB/s,  743.3 MB/s
Benchmarking /home/cyan/dev/bench/dirCalgary/book2 
 1#book2             :    610856 ->    228144 (x2.678),  229.4 MB/s, 1025.0 MB/s
 2#book2             :    610856 ->    215190 (x2.839),  182.2 MB/s,  859.0 MB/s
 3#book2             :    610856 ->    203784 (x2.998),  152.8 MB/s,  854.1 MB/s

[terrelln]

Now that the benchmark mode uses the same default nb of threads as normal CLI compression, it's no longer obvious how many threads are being used in benchmark

Is this the desired behavior? I'd expect that benchmark mode would use single threaded compression by default.

[Cyan4973]

Is this the desired behavior? I'd expect that benchmark mode would use single threaded compression by default.

We discussed this behavior with @daniellerozenblit when she introduced MT by default in zstd,
and the consensus was that it made more sense for the benchmark to feature the same default as CLI, because casual users will generally expect that they are measuring the same thing.

That's definitely something that can still be discussed if you wish to.

In the meantime, this PR just displays this information to make sure users understand what is being measured, so the functionality remains valid whatever the decision is regarding default threads behavior.